What is a core banking system?

A core banking system is the software that holds a bank's accounts and ledger and processes the transactions on them: deposits, withdrawals, payments, interest, fees and loan balances. It is the bank's system of record for who holds what. Everything else, from the mobile app to the CRM, reads from it or writes to it.

Why does the core matter once AI agents act?

It is where an agent's action becomes real. A note in a CRM can be edited. A posting to the ledger moves money, changes a balance the customer sees, and feeds regulatory reporting. An agent that can write to the core is acting at the heart of the bank, and its access deserves the same care as a person with posting rights.

The core is also the strongest source of facts an agent can have. A balance read from the ledger is issued by the system of record itself. A balance quoted in a customer's email, or in a screenshot, is a claim from outside. An agent deciding whether a fee is due should rely on the first and treat the second as something to check.

How does an agent work with a core banking system?

• Read access through the bank's interfaces, limited to the fields the agent's role needs.
• Write access only through defined transactions, such as "update address" or "apply fee waiver", never through the database directly.
• Limits on each write, so a fee waiver above a set amount is held for a person.
• Identity: every change is tagged with the agent that proposed it and the person who approved it, if anyone did.
• Maker-checker on postings that move money, so the agent can prepare but not complete them alone.

The practical steps are in how to give an AI agent safe access to your core banking system. Changing customer data has its own risks, covered in how to let an AI agent update customer records in your core banking system.

What does that look like in an example?

A microfinance bank's servicing agent handles address changes. It reads the customer record from the core, checks the uploaded utility bill against the name on the account, and updates the address through the defined transaction at depth 3. The same customer asks for a 40 late fee to be reversed. That is a posting, so the agent drafts it and holds it for a branch officer. The bank's limits, not the agent, decided which request went through and which waited.

How is a core banking system different from open banking?

Open banking is a way for authorised third parties to read account data or start payments, with the customer's consent, through interfaces the bank exposes. The core banking system is the bank's internal ledger behind those interfaces. Open banking is a door; the core is the vault. An outside agent using open banking never touches the core directly; an internal agent may, which is why its permissions need closer design.

Which related terms should you know?

• General ledger: the bank's own accounting books, fed by the core but separate from customer accounts.
• Online banking: the customer-facing channel. Resetting a customer's access to it is a security action, discussed in can an AI agent reset online banking access?

Where Pryme fits

Through the Pryme Context Protocol, facts read from the core banking ledger arrive stamped as sealed, the strongest provenance class. Writes are checked against the agent's role, certificate, sources and the bank's own limits before they reach the core, and each verdict is filed. The broader limits are in AI agents in banking.